Stream: t-compiler/wg-pipelining

Topic: implementation details


eddyb (Apr 30 2019 at 15:35, on Zulip):

it seems <https://github.com/rust-lang/rust/pull/60006> was merged without my proposed change (or an argument against) but I think GitHub may be to blame for my comment being lost, heh (and I didn't see the further notifications, oops)

eddyb (Apr 30 2019 at 15:36, on Zulip):

@nnethercote @Alex Crichton would you be against me preparing a PR that emits an uniform JSON message on any file output, as a sort of "output progress"?

eddyb (Apr 30 2019 at 15:37, on Zulip):

oh god I'm trying to load the 19 hidden comments on the PR and it's killing my laptop

Alex Crichton (Apr 30 2019 at 15:39, on Zulip):

@eddyb the thinking on my end at least was that this is all unstable and forwards compatible with that

Alex Crichton (Apr 30 2019 at 15:39, on Zulip):

in the sense that the goal here is to get data about speeding up builds thorugh pipelining

Alex Crichton (Apr 30 2019 at 15:39, on Zulip):

rather than designing a general system of notifications coming out of the compiler

Alex Crichton (Apr 30 2019 at 15:39, on Zulip):

(but leaving room for that to exist of course)

eddyb (Apr 30 2019 at 15:40, on Zulip):

my point is that we have a bad history of accidentally stabilizing some temporary implementation :P

eddyb (Apr 30 2019 at 15:41, on Zulip):

so I'd rather have the dumbest/simplest/most general thing here, and let Cargo use it for pipelining

Alex Crichton (Apr 30 2019 at 15:41, on Zulip):

perhaps, but I also feel like we have a bad history of getting hung up on driving to a "pure" solution while obvious wins are left on the table

eddyb (Apr 30 2019 at 15:41, on Zulip):

or sccache for picking out files, etc.

Alex Crichton (Apr 30 2019 at 15:41, on Zulip):

the most simple thing for now is what's already there?

Alex Crichton (Apr 30 2019 at 15:41, on Zulip):

(which we haven't committed to stabilizing at all yet)

eddyb (Apr 30 2019 at 15:42, on Zulip):

I suspect it would be easier to replace all fs::rename calls, but I guess I can wait until I've actually attempted it, to say that (and not risk eating my own words)

eddyb (Apr 30 2019 at 15:44, on Zulip):

moving metadata emission earlier would be more of an endeavor, I suspect

eddyb (Apr 30 2019 at 15:59, on Zulip):

@nnethercote how was this comment resolved? I see no reply to it https://github.com/rust-lang/rust/pull/60006#discussion_r278147115

nnethercote (Apr 30 2019 at 22:29, on Zulip):

@eddyb I have vague YAGNI concerns about doing it before there's an explicit need, but I'm not against it.

nnethercote (Apr 30 2019 at 22:33, on Zulip):

@eddyb: ugh, I wrote a reply but must have failed to submit it. I have rewritten it now.

nnethercote (Apr 30 2019 at 22:46, on Zulip):

hmm, I probably clicked "resolve conversation", thinking that would post the comment I wrote

eddyb (May 01 2019 at 13:17, on Zulip):

ouch!

eddyb (May 02 2019 at 01:06, on Zulip):

so I finally has a chance to take a look at this: the original PR removed some out_filenames.push(out_filename); lines! that's pretty much where I'd expect something to be emitted

eddyb (May 02 2019 at 01:07, on Zulip):

with something like --crate-type=rlib,dylib, you might spend a little while linking one then the other, so they would be separated in time, I think?

eddyb (May 02 2019 at 01:10, on Zulip):

and it's actually only 1 other call site

eddyb (May 02 2019 at 02:57, on Zulip):

okay I opened https://github.com/rust-lang/rust/pull/60464

eddyb (May 02 2019 at 02:57, on Zulip):

now I can forget about it :P

Last update: Nov 15 2019 at 09:40UTC